A Smart Early Warning and Credit Assessment Method for Vehicle Transaction Risks Based on Big Data Analysis

Abstract

This invention discloses a method for intelligent early warning and credit assessment of vehicle transaction risks based on big data analysis, comprising: S1, collecting a set of original transaction data; S2, constructing a heterogeneous graph structure based on the original transaction data set; S3, dividing the heterogeneous graph structure into a sequence of graph snapshots according to fixed time windows, generating a time evolution graph sequence with time evolution characteristics; S4, constructing low-dimensional feature representations of user nodes and transaction event nodes under each time window; S5, constructing a causal path graph based on the temporal order and behavioral dependencies of vehicle transaction record data, and extracting high-frequency transaction causal chains from it; S6, jointly modeling the low-dimensional feature representations of user nodes and transaction event nodes with the high-frequency transaction causal chains; and S7, outputting the risk level label of the target transaction and the corresponding user's credit score result. This invention has the advantages of strong structural expression ability, clear causal logic, and high accuracy of risk assessment.

Description

Technical Field

[0001] This invention relates to the field of intelligent early warning technology, and in particular to an intelligent early warning and credit assessment method for vehicle transaction risks based on big data analysis. Background Technology

[0002] With the rapid growth of car ownership and the continuous expansion of the used car market, vehicle transaction scenarios face highly complex participant structures, diverse transaction behavior characteristics, and a market environment with high information asymmetry. In the absence of a unified credit assessment system and dynamic risk identification mechanism, how to accurately model the behavior patterns of transaction participants, provide early warnings of transaction risks, and scientifically evaluate user credit has become one of the core issues in the field of smart transportation and intelligent transaction supervision.

[0003] In existing technologies, vehicle transaction risk assessment mainly relies on static rule engines or expert experience modeling methods. The core approach is often a scoring system combining behavioral feature extraction and threshold judgment. This approach suffers from significant shortcomings, including a single assessment dimension, slow risk feature updates, and a lack of interpretability in the scoring mechanism. Particularly when facing heterogeneous transaction graph structures with complex participant identities and densely intersecting behavioral chains, traditional models cannot effectively capture the dynamic evolutionary relationships between multi-source data. This leads to decreased transaction risk identification capabilities, distorted credit scoring results, and assessment biases such as "high score but high risk" or "low score but stable reputation." Furthermore, while some studies have attempted to introduce machine learning and deep models for risk prediction, most only model at the feature level, lacking the ability to model the evolutionary process of the transaction structure and failing to integrate causal logical explanation mechanisms. These models remain at the "black box prediction" stage, unable to support the practical needs of intelligent regulatory scenarios for traceable transactions, interpretable risks, and interventionist decisions.

[0004] In terms of decision support for risk scoring results, existing technologies generally adopt a fixed scoring interval to divide levels, which does not adequately consider the impact of different user behavior sequences, transaction paths and historical risk records. It lacks a flexible control mechanism that combines dynamic structural changes and risk propagation paths, resulting in poor real-time performance and ability to express differences in credit scores, which is not conducive to achieving multi-dimensional risk perception and dynamic credit evolution control in complex transaction systems.

[0005] In summary, there is an urgent need for a vehicle transaction risk intelligent early warning and credit assessment method based on big data analysis, in order to achieve deep integration of structural evolution characteristics modeling, risk causal path extraction and joint risk and credit assessment in the vehicle transaction process, and effectively solve the core problems of static assessment models, lagging risk identification and unexplainable scoring mechanisms in existing technologies. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent early warning and credit assessment method for vehicle transaction risks based on big data analysis. This invention integrates a structurally adaptive heterogeneous graph neural network with a credit behavior causal reasoning engine, constructs a time evolution graph modeling, node type-aware embedding, and high-frequency transaction causal chain extraction mechanism, and achieves quantitative assessment of transaction risk level and user credit score through joint scoring graph generation and risk correlation function. It also realizes dynamic risk identification and interpretable credit modeling processing under complex transaction structures, and has the advantages of strong structural expression ability, clear causal logic, and high accuracy of risk assessment.

[0007] According to an embodiment of the present invention, a method for intelligent early warning and credit assessment of vehicle transaction risks based on big data analysis includes the following steps:

[0008] S1. Collect vehicle transaction record data, user behavior trajectory data, third-party credit data and social interaction data, and perform identification normalization and time sequence alignment to generate the original transaction data set;

[0009] S2. Construct a heterogeneous graph structure based on the original transaction data set;

[0010] S3. Divide the heterogeneous graph structure into a sequence of graph snapshots according to a fixed time window, and perform adaptive structural reconstruction on the graph snapshots in each sequence based on the node addition rate and edge change rate to generate a time evolution graph sequence with time evolution characteristics.

[0011] S4. For each snapshot of the time evolution graph sequence, a graph neural network is used to perform node embedding calculation to generate low-dimensional feature representations of user nodes and transaction event nodes under each time window.

[0012] S5. Based on the original transaction data set and time evolution sequence, construct a causal path graph according to the time sequence and behavioral dependencies of vehicle transaction record data, and extract high-frequency transaction causal chains from it;

[0013] S6. Jointly model the low-dimensional feature representations of user nodes and transaction event nodes with the high-frequency transaction causal chain to generate a transaction scoring graph, and define a risk correlation function to calculate the vehicle transaction risk coefficient and user credit score.

[0014] S7. Based on the transaction rating chart, output the risk level label of the target transaction and the corresponding user's credit score result.

[0015] Optionally, the original transaction data set includes collected vehicle transaction record data, user behavior trajectory data, third-party credit data, and social interaction data, and performs identification normalization and time-series alignment. Among them, vehicle transaction record data is used to identify the transaction behavior relationship between users and vehicles, user behavior trajectory data is used to identify the account operation similarity relationship between different user accounts, and social interaction data is used to mine the behavior co-occurrence relationship between users.

[0016] Optionally, the construction of the heterogeneous graph structure includes: constructing user nodes, vehicle nodes, transaction event nodes, and social nodes respectively using user entities, vehicle entities, transaction event entities, and social entities extracted from the original transaction data set; generating transaction behavior relationship edges between user nodes and vehicle nodes based on the mapping relationship between user identifiers and vehicle identifiers contained in the vehicle transaction record data; constructing account operation similarity relationship edges between different user accounts with similar account operation characteristics in the user behavior trajectory data; and generating social co-occurrence relationship edges between user pairs that participate in transaction behavior together based on message interaction behavior in the social interaction data, thus forming a heterogeneous graph structure containing multiple types of nodes and multiple types of relationship edges.

[0017] Optionally, S3 specifically includes:

[0018] S31. Divide the heterogeneous graph structure into time segments according to fixed time windows to obtain a sequence of graph snapshots. Among them, G t This represents the snapshot of the graph corresponding to the t-th time window, where n represents the total number of time windows;

[0019] S32, For each graph snapshot G t Calculate the node addition rate α t The rate of change of the edge β t :

[0020]

[0021] Among them, V t and E t Representing the snapshot G respectively t The set of nodes and the set of edges, V t-1 With E t-1 These represent the set of nodes and the set of edges of the previous time window snapshot, respectively. The symbol \ indicates the set difference operation, and the symbol △ indicates the symmetric difference of sets.

[0022] S33. When the following adaptive reconstruction condition is met:

[0023] α t >θ v ∪β t >θ e ;

[0024] Then for the snapshot G t Perform a structural update and update the graph snapshot G. t The adjacency matrix and node attribute representation, where θ v With θ e These are the thresholds for node addition rate and edge change rate, respectively.

[0025] S34. Arrange all the graph snapshots that have undergone structural updates in chronological order to form a time evolution graph sequence. Among them, G' t Let m represent the graph snapshot after adaptive reconstruction, and m represent the number of updated graph snapshots, satisfying m≤n.

[0026] Optionally, S33 specifically includes:

[0027] S331. For a graph snapshot G that satisfies the adaptive reconstruction condition... t All nodes v i Perform structural disturbance detection and compare the current neighbor set. Snapshot G from the previous image t-1 Neighbor set Constructing a set of perturbed neighbors

[0028] S332, For snapshot G t Each node v i Based on the current node representation With the set of disturbed neighbor nodes The structure-aware weighted aggregation result is used to calculate the node representation after the structure update. Its expression is:

[0029]

[0030] Where γ∈[0,1] is the structural perturbation fusion factor. Indicates the disturbance of neighbor node v j At time t, the node is represented as follows: For node v j For node v i The perturbation structure affects the weights, satisfying:

[0031] S333. Based on the node representation update result, reconstruct the graph snapshot G. t The adjacency matrix and edge set are used to retain the edge perturbation threshold θ whose change intensity exceeds the edge perturbation threshold. s Given the edge structure, generate a snapshot G' of the updated graph. t .

[0032] Optionally, S4 specifically includes:

[0033] S41. For each of the time evolution graph snapshots G′ after adaptive structural reconstruction in the time evolution graph sequence. t User nodes and transaction event nodes are embedded and initialized separately, and initial representation vectors are set according to node types.

[0034] S42. Before each propagation layer l begins, based on the distribution structure of heterogeneous node types in the graph snapshot, execute the type constraint sampling strategy for neighbor nodes to construct the type-aware neighbor set for each node in the propagation layer.

[0035] S43. For each node v i Based on type-aware neighbor sets Construct the propagation input and collect the representations of neighboring nodes in the previous layer. And retain node type information;

[0036] S44. Update node representations using a node type-aware aggregation mechanism:

[0037]

[0038] in, Represents node v in the l-th layer i The representation vector, Represents node v j The representation vector in the previous propagation layer l-1, Represents node v i In the set of neighboring nodes in the current graph snapshot, σ(·) represents a non-linear activation function, and MEAN(·) represents an aggregation function that performs an element-wise averaging operation on all input vectors in the set. This indicates that the node v i The trainable weight matrix corresponding to the node type φ(i) at the l-th layer, where φ(i) is the node type mapping function;

[0039] S45, in the snapshot G′ t In each propagation layer l that performs graph neural network embedding computation, the node representation dimension and activation function type of the propagation layer are set, and the corresponding type propagation weight matrix in the propagation layer is configured for different types of graph nodes to control the representation update method of each type of graph node in the propagation layer.

[0040] S46. After all propagation layers have been executed, extract the final representations of the user node and the transaction event node respectively. As a low-dimensional semantic representation of user nodes and transaction event nodes.

[0041] Optionally, S42 specifically includes:

[0042] S421. In each graph snapshot G′ after adaptive structural reconstruction t In the middle, for each target graph node v i Obtain the type φ(i) of the node and extract the set of neighboring nodes that have connecting edges with the node in the graph. As a set of candidate neighbors;

[0043] S422. Construct a type constraint matrix between node types. Where K is the total number of node types in the graph, Θ ab ∈[0,1] indicates that graph nodes of type a are allowed to sample graph nodes of type b;

[0044] S423, Based on the target graph node v i The type φ(i) = a, for the set of candidate neighbor nodes Graph node v of type b j Assign sampling probability p ij ;

[0045] S424, Based on the sampling probability p ij For candidate neighbor set Perform a sampling operation to obtain the target graph node v i Type-aware neighbor set in the current propagation layer l The number of sampling nodes is controlled by the hyperparameter d;

[0046] S425. To improve the stability of the sampling process, multiple rounds of resampling are performed and the sampling frequency of candidate nodes is counted. If the frequency deviation exceeds a set threshold ∈, the sampling probability distribution p of the candidate neighbors is renormalized. ij .

[0047] Optionally, S5 specifically includes:

[0048] S51, From the time evolution sequence Extract the transaction event nodes that occur within each time window, and combine them with the timestamps from the original transaction dataset to construct a global event sequence {e1, e2, ..., e...} for all transaction events in chronological order. n}, where e i This represents the i-th transaction event node, and n represents the total number of events;

[0049] S52, in each graph snapshot G' after adaptive structural reconstruction t In the graph structure, based on the structural connections and behavioral attribute dependencies between transaction event nodes, it is determined whether event pairs in the global event sequence satisfy the causal triggering condition. If event e i Prior to e in time j And in G't If structured behavioral interactions exist, then causal edges (e) are added to the causal path graph. i →e j );

[0050] S53, Adding causal edges (e i →e j At the same time, causal strength weights w are further assigned to the causal edges. ij w ij ∈[0,1]:

[0051] w ij =α·freq(e i ,e j )+β·risk(e i );

[0052] Among them, freq(e i ,e j ) represents an event pair (e i ,e j The historical co-occurrence frequency of causal edges identified in the time evolution graph sequence, risk(e) i ) represents event e i The risk value corresponding to the transaction behavior, where α and β are weighting balance coefficients.

[0053] S54. Calculate the cumulative weight values ​​of all causal edges in the causal path graph that are assigned causal strength weights, set a causal strength threshold, and extract causal edges whose cumulative weight values ​​are not lower than the causal strength threshold as high-frequency trading causal chains.

[0054] Optionally, S6 specifically includes:

[0055] S61. The final embedded representation of user nodes and transaction event nodes. The statistical feature vector c formed by the causal edge frequency and edge weight information in the associated high-frequency trading causal chain. i Perform joint modeling to generate joint feature representation z of graph nodes in the transaction scoring graph. i ;

[0056] S62, Based on joint feature representation z i Construct a transaction rating graph and calculate the value of each edge (v) in the transaction rating graph. i ,v j Risk association score r ij :

[0057]

[0058] Where, r ij ∈[0,1] represents the degree of risk association between graph node pairs, zi With z j Representing graph node v i With v j The joint feature representation, M is the trainable scoring weight matrix, w ij λ represents the causal edge weights between node pairs in the causal path graph, λ∈[0,1] is the causal weight adjustment coefficient, and σ(·) is the Sigmoid function.

[0059] S63. Based on the risk correlation scores between user nodes and transaction event nodes in the rating graph, calculate the vehicle transaction risk coefficient and user credit score respectively, and output the risk level label of each transaction behavior and the credit score result of each user.

[0060] Optionally, S63 specifically includes:

[0061] S631. Based on the risk correlation score r between all user nodes and the target transaction event node in the transaction scoring graph. ij Calculate transaction event node v j Vehicle transaction risk coefficient R j :

[0062]

[0063] in, Indicates the transaction event node v j There exists a set of all user nodes connected by the rating edge, r. ij Represents user node v i With transaction event node v j Risk association score, R j This represents the risk coefficient of a transaction event, with a value ranging from 0 to 1.

[0064] S632. Based on the risk correlation score r between all transaction event nodes and target user nodes in the transaction scoring graph. ij Calculate user node v i User credit score C i :

[0065]

[0066] in, Indicates the relationship with user node v i There exists a set C of all transaction event nodes connected by a rating edge. i This represents the user's credit score, ranging from 0 to 1.

[0067] S633, Based on the vehicle transaction risk coefficient R j And user credit score C iThe risk level label is output for each transaction node, and the credit score is output for each user node. The risk level label is divided into three levels. When the risk coefficient R... j A value greater than 0.75 is marked as a high-risk level. j A risk level greater than 0.4 and less than or equal to 0.75 is designated as medium risk. j A value less than or equal to 0.4 is marked as a low-risk level.

[0068] The beneficial effects of this invention are:

[0069] (1) This invention introduces a structural adaptive reconstruction mechanism in the process of heterogeneous graph modeling. Based on the node addition rate and edge change rate, a process for detecting graph snapshot structure disturbance and generating time evolution graph sequence is constructed. This can dynamically reflect the structural change trend of vehicle transaction behavior in the time dimension, effectively overcome the limitation of existing methods that rely on static graph structure modeling, improve the ability to capture the evolution characteristics of complex transaction behavior chains over time, and provide structural foundation support for subsequent time-series risk identification and credit evolution assessment.

[0070] (2) This invention constructs a type-aware multi-layer graph neural network propagation mechanism, and combines node initialization, neighbor type sampling and heterogeneous weight aggregation strategies to generate low-dimensional semantic embedding representations of user nodes and transaction event nodes. It has the characteristics of strong semantic discriminability and high structural context consistency, and breaks through the bottleneck of semantic ambiguity and poor representation consistency of traditional embedding models in heterogeneous graphs. It provides a highly expressive feature basis for risk association modeling and scoring between nodes of multiple types of graphs.

[0071] (3) Based on risk identification, this invention integrates high-frequency transaction causal chain and joint scoring graph modeling method to establish a dual-indicator quantitative system of transaction risk coefficient and user credit score. Based on the causal edge frequency and transaction risk contribution, it constructs an interpretable risk propagation structure, which significantly improves the interpretability and traceability of risk assessment results. It breaks through the black box limitation of existing models that only predict results based on feature fitting, and provides a reliable basis for transaction risk classification, credit labeling and regulatory intervention. Attached Figure Description

[0072] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0073] Figure 1 This is an overall flowchart of a vehicle transaction risk intelligent early warning and credit assessment method based on big data analysis proposed in this invention;

[0074] Figure 2This is a schematic diagram of the processing flow of constructing a time evolution graph sequence based on a structure adaptive evolution mechanism, which is a method for intelligent early warning and credit assessment of vehicle transaction risks based on big data analysis proposed in this invention.

[0075] Figure 3 This is a schematic diagram illustrating the structure of a vehicle transaction risk intelligent early warning and credit assessment method based on big data analysis proposed in this invention, which integrates a type-aware graph neural network with a high-frequency transaction causal chain to jointly generate risk labels and credit scores. Detailed Implementation

[0076] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0077] refer to Figures 1-3 A method for intelligent early warning and credit assessment of vehicle transaction risks based on big data analysis includes the following steps:

[0078] S1. Collect vehicle transaction record data, user behavior trajectory data, third-party credit data and social interaction data, and perform identification normalization and time sequence alignment to generate the original transaction data set;

[0079] S2. Construct a heterogeneous graph structure based on the original transaction data set;

[0080] S3. Divide the heterogeneous graph structure into a sequence of graph snapshots according to a fixed time window, and perform adaptive structural reconstruction on the graph snapshots in each sequence based on the node addition rate and edge change rate to generate a time evolution graph sequence with time evolution characteristics.

[0081] S4. For each snapshot of the time evolution graph sequence, a graph neural network is used to perform node embedding calculation to generate low-dimensional feature representations of user nodes and transaction event nodes under each time window.

[0082] S5. Based on the original transaction data set and time evolution sequence, construct a causal path graph according to the time sequence and behavioral dependencies of vehicle transaction record data, and extract high-frequency transaction causal chains from it;

[0083] S6. Jointly model the low-dimensional feature representations of user nodes and transaction event nodes with the high-frequency transaction causal chain to generate a transaction scoring graph, and define a risk correlation function to calculate the vehicle transaction risk coefficient and user credit score.

[0084] S7. Based on the transaction rating chart, output the risk level label of the target transaction and the corresponding user's credit score result.

[0085] By integrating and aligning multi-source data (vehicle transaction records, user behavior trajectory data, third-party credit data, and social interaction data) with time series in this embodiment of the invention, a unified abstract representation of the original transaction behavior can be achieved. Combining a structure-adaptive heterogeneous graph construction mechanism with the generation of time evolution graph snapshot sequences, the dynamic behavioral evolution patterns among transaction participants are effectively captured. Semantic separation and low-dimensional embedding are achieved through a node-type-aware graph neural network propagation process, improving the class distinguishability and structural stability of node feature expressions. Based on this, a causal path graph constructed from the transaction record time sequence and structural interaction relationship, combined with a high-frequency causal edge extraction mechanism and a causal weight design strategy, forms an interpretable risk path representation. Finally, through the transaction rating graph generation process, transaction risk and user credit characteristics are jointly modeled, outputting vehicle transaction risk level and user credit score results. This achieves a closed-loop linkage processing from risk prediction and causal modeling to rating interpretation, significantly enhancing the model's dynamic expressiveness, causal interpretability, and practicality of joint evaluation in a multi-source heterogeneous data environment.

[0086] In this embodiment, the original transaction data set includes collected vehicle transaction record data, user behavior trajectory data, third-party credit data, and social interaction data, and performs identification normalization and time sequence alignment. Among them, vehicle transaction record data is used to identify the transaction behavior relationship between users and vehicles, user behavior trajectory data is used to identify the account operation similarity relationship between different user accounts, and social interaction data is used to mine the behavior co-occurrence relationship between users.

[0087] This invention establishes a multi-dimensional interactive relationship structure oriented towards users and vehicles, and users and users, by normalizing and integrating vehicle transaction record data, user behavior trajectory data, third-party credit data, and social interaction data. This effectively realizes a multi-dimensional behavioral network expression from transaction behavior, account operation patterns to social behavior associations. Among them, vehicle transaction record data is used to explicitly define the transaction behavior relationship between users and vehicles, user behavior trajectory data is used to characterize the operational similarity between different user accounts, and social interaction data is used to discover potential behavioral co-occurrence relationships between users in the social chain. This original data fusion mechanism provides a high-quality and structurally clear multi-relationship foundation for the subsequent construction of heterogeneous graph structures, significantly improving the system's ability to identify risk nodes and potential abnormal patterns in vehicle transaction behavior.

[0088] In this embodiment, the construction of the heterogeneous graph structure includes: constructing user nodes, vehicle nodes, transaction event nodes, and social nodes respectively using user entities, vehicle entities, transaction event entities, and social entities extracted from the original transaction data set; generating transaction behavior relationship edges between user nodes and vehicle nodes based on the mapping relationship between user identifiers and vehicle identifiers contained in the vehicle transaction record data; constructing account operation similarity relationship edges between different user accounts with similar account operation characteristics in the user behavior trajectory data; and generating social co-occurrence relationship edges between user pairs that participate in transaction behavior and have message interaction behavior in the social interaction data, thus forming a heterogeneous graph structure containing multiple types of nodes and multiple types of relationship edges.

[0089] This invention introduces multiple types of graph node entities, such as user nodes, vehicle nodes, transaction event nodes, and social nodes, when constructing a heterogeneous graph structure. Based on the transaction mapping relationship between users and vehicles, the similarity relationship of account operation behavior, and the co-occurrence relationship of social interaction among users, transaction behavior relationship edges, account similarity relationship edges, and social co-occurrence relationship edges are constructed respectively, forming a heterogeneous graph model with a clear structure and semantics. This structure effectively supports the modeling of complex behaviors of vehicle transaction participants in different dimensions, enabling the system to shift from static entity description to relationship-driven modeling. It lays a multi-graph structure foundation for subsequent structural evolution modeling and graph embedding propagation, significantly enhancing the model's ability to express complex transaction behavior graphs and preserve structural context.

[0090] In this embodiment, S3 specifically includes:

[0091] S31. Divide the heterogeneous graph structure into time segments according to fixed time windows to obtain a sequence of graph snapshots. Among them, G t This represents the snapshot of the graph corresponding to the t-th time window, where n represents the total number of time windows;

[0092] S32, For each graph snapshot G t Calculate the node addition rate α t The rate of change of the edge β t :

[0093]

[0094] Among them, V t and E t Representing the snapshot G respectively t The set of nodes and the set of edges, V t-1 With E t-1 These represent the set of nodes and the set of edges of the previous time window snapshot, respectively. The symbol \ indicates the set difference operation, and the symbol △ indicates the symmetric difference of sets.

[0095] By using formulas for node addition rate and edge change rate, the structural changes between each snapshot of the graph in the time evolution graph sequence and the previous time window can be quantitatively expressed. This allows the system to sensitively perceive the node addition trend and dynamic changes in edge connections at the graph structure level. Based on the calculation method of set difference and normalized ratio, this formula constructs a mechanism for measuring the relative change amplitude of structural disturbances. It can avoid absolute offset errors caused by different total number of nodes or edge density, effectively improving the stability and universality of graph reconstruction judgment criteria. It provides clear, controllable, and adjustable dynamic triggering criteria for subsequent adaptive structural reconstruction and disturbance detection, enhancing the scientificity and feasibility of structural change decisions in time evolution modeling.

[0096] S33. When the following adaptive reconstruction condition is met:

[0097] α t >θ v ∪β t >θ e ;

[0098] Then for the snapshot G t Perform a structural update and update the graph snapshot G. t The adjacency matrix and node attribute representation, where θ v With θ e These are the thresholds for node addition rate and edge change rate, respectively.

[0099] S34. Arrange all the graph snapshots that have undergone structural updates in chronological order to form a time evolution graph sequence. Among them, G' t Let m represent the graph snapshot after adaptive reconstruction, and m represent the number of updated graph snapshots, satisfying m≤n.

[0100] In this embodiment, S33 specifically includes:

[0101] S331. For a graph snapshot G that satisfies the adaptive reconstruction condition... t All nodes v i Perform structural disturbance detection and compare the current neighbor set. Snapshot G from the previous image t-1 Neighbor set Constructing a set of perturbed neighbors

[0102] S332, For snapshot G t Each node v i Based on the current node representation With the set of disturbed neighbor nodes The structure-aware weighted aggregation result is used to calculate the node representation after the structure update. Its expression is:

[0103]

[0104] Where γ∈[0,1] is the structural perturbation fusion factor. Indicates the disturbance of neighbor node v j At time t, the node is represented as follows: For node v j For node v i The perturbation structure affects the weights, satisfying:

[0105] By employing a node update formula based on weighted aggregation of structurally perturbed neighbors, a refined model of the dynamic adjustment mechanism of node representations in graph snapshots under structural changes can be achieved. This formula integrates the representation of the current node itself with the representation of the perturbed neighbor set, and controls the influence weights of the two through a structural perturbed fusion factor. At the same time, normalized perturbed influence weights are used to ensure the stability and controllability of the aggregation process, thereby achieving smooth evolution and information absorption of nodes after structural perturbation. This design effectively improves the coherence and sensitivity of node representation updates under graph structural changes, and can introduce new neighbor semantics while maintaining the original structural information. It provides a stable node-level update foundation for modeling the structural continuity of graph snapshots during the evolution process, and enhances the robustness of graph neural networks in heterogeneous dynamic graph environments.

[0106] S333. Based on the node representation update result, reconstruct the graph snapshot G. t The adjacency matrix and edge set are used to retain the edge perturbation threshold θ whose change intensity exceeds the edge perturbation threshold. s Given the edge structure, generate a snapshot G' of the updated graph. t .

[0107] This invention introduces a structural perturbation detection mechanism based on node addition rate and edge change rate during graph structure modeling. It employs a time-slicing strategy to divide heterogeneous graphs into a sequence of graph snapshots, and performs perturbation threshold determination and structural update operations on each snapshot, achieving dynamic adaptive reconstruction of the graph structure over time. Furthermore, by performing neighbor set change detection on graph snapshot nodes that meet the reconstruction conditions, and performing structure-aware weighted aggregation based on the perturbed neighbor set, it generates an embedded representation of the updated node structure. Then, based on the updated node representation, it reconstructs the adjacency matrix and edge set, ultimately generating a time-evolved graph sequence containing graph evolution features. This mechanism effectively improves the graph neural network's ability to model the evolutionary characteristics of vehicle transaction behavior over time, enhances the adaptability of node representations to graph structure changes, and provides a dynamically consistent graph representation foundation for subsequent semantic embedding and causal path extraction.

[0108] In this embodiment, S4 specifically includes:

[0109] S41. For each of the time evolution graph snapshots G′ after adaptive structural reconstruction in the time evolution graph sequence. t User nodes and transaction event nodes are embedded and initialized separately, and initial representation vectors are set according to node types.

[0110] S42. Before each propagation layer l begins, based on the distribution structure of heterogeneous node types in the graph snapshot, execute the type constraint sampling strategy for neighbor nodes to construct the type-aware neighbor set for each node in the propagation layer.

[0111] S43. For each node v i Based on type-aware neighbor sets Construct the propagation input and collect the representations of neighboring nodes in the previous layer. And retain node type information;

[0112] S44. Update node representations using a node type-aware aggregation mechanism:

[0113]

[0114] in, Represents node v in the l-th layer i The representation vector, Represents node v j The representation vector in the previous propagation layer l-1, Represents node v i In the set of neighboring nodes in the current graph snapshot, σ(·) represents a non-linear activation function, and MEAN(·) represents an aggregation function that performs an element-wise averaging operation on all input vectors in the set. This indicates that the node v i The trainable weight matrix corresponding to the node type φ(i) at the l-th layer, where φ(i) is the node type mapping function;

[0115] The node representation update mechanism achieves semantic difference modeling of node representations in heterogeneous graphs by averaging the features of neighboring nodes of the target node in the current graph snapshot and jointly updating them with the target node's own historical features. It also introduces a node type-aware weight matrix, allowing different types of nodes to be modeled using corresponding independent parameters during propagation. This avoids the semantic confusion problem caused by homogeneous aggregation in heterogeneous structures, improves the accuracy of node feature expression, and enhances the model's generalization ability in multi-type node distribution structures. This provides a stable and distinguishable node embedding representation foundation for subsequent scoring and risk assessment.

[0116] S45, in the snapshot G′ tIn each propagation layer l that performs graph neural network embedding computation, the node representation dimension and activation function type of the propagation layer are set, and the corresponding type propagation weight matrix in the propagation layer is configured for different types of graph nodes to control the representation update method of each type of graph node in the propagation layer.

[0117] S46. After all propagation layers have been executed, extract the final representations of the user node and the transaction event node respectively. As a low-dimensional semantic representation of user nodes and transaction event nodes.

[0118] In this embodiment, S42 specifically includes:

[0119] S421. In each graph snapshot G′ after adaptive structural reconstruction t In the middle, for each target graph node v i Obtain the type φ(i) of the node and extract the set of neighboring nodes that have connecting edges with the node in the graph. As a set of candidate neighbors;

[0120] S422. Construct a type constraint matrix between node types. Where K is the total number of node types in the graph, Θ ab ∈[0,1] indicates that graph nodes of type a are allowed to sample graph nodes of type b;

[0121] S423, Based on the target graph node v i The type φ(i) = a, for the set of candidate neighbor nodes Graph node v of type b j Assign sampling probability p ij ;

[0122] S424, Based on the sampling probability p ij For candidate neighbor set Perform a sampling operation to obtain the target graph node v i Type-aware neighbor set in the current propagation layer l The number of sampling nodes is controlled by the hyperparameter d;

[0123] S425. To improve the stability of the sampling process, multiple rounds of resampling are performed and the sampling frequency of candidate nodes is counted. If the frequency deviation exceeds a set threshold ∈, the sampling probability distribution p of the candidate neighbors is renormalized. ij .

[0124] This invention introduces a type-aware graph neural network propagation mechanism into the graph node embedding representation generation process, constructing a multi-layer embedding computation flow from node initialization, type constraint sampling to weight aggregation and updating. This enables deep modeling of the behavioral structural context of user nodes and transaction event nodes in heterogeneous graphs. By constructing a type-aware neighbor set for each propagation layer, semantic constraints on the propagation adjacency range are achieved. Furthermore, based on the type association matrix, cross-type sampling probabilities are calculated to generate a high-fidelity structural context neighbor set, and sampling bias verification is performed in multiple rounds of sampling, enhancing the stability of adjacency modeling and type separation capability. In addition, a node type-aware weight matrix is ​​introduced into the node propagation representation update, constructing a type-based parameter sharing and non-sharing strategy to achieve differentiated control of feature representation for different node types. Finally, low-dimensional semantic representations of user nodes and transaction event nodes are output, providing a structurally stable and semantically clear graph embedding representation foundation for subsequent joint modeling, improving the system's embedding representation capability and generalization performance in complex heterogeneous graph environments.

[0125] In this embodiment, S5 specifically includes:

[0126] S51, From the time evolution sequence Extract the transaction event nodes that occur within each time window, and combine them with the timestamps from the original transaction dataset to construct a global event sequence {e1, e2, ..., e...} for all transaction events in chronological order. n}, where e i This represents the i-th transaction event node, and n represents the total number of events;

[0127] S52, in each graph snapshot G' after adaptive structural reconstruction t In the graph structure, based on the structural connections and behavioral attribute dependencies between transaction event nodes, it is determined whether event pairs in the global event sequence satisfy the causal triggering condition. If event e i Prior to e in time j And in G' t If structured behavioral interactions exist, then causal edges (e) are added to the causal path graph. i →e j );

[0128] S53, Adding causal edges (e i →e j At the same time, causal strength weights w are further assigned to the causal edges. ij w ij ∈[0,1]:

[0129] w ij =α·freq(e i ,e j )+β·risk(ei );

[0130] Among them, freq(e i ,e j ) represents an event pair (e i ,e j The historical co-occurrence frequency of causal edges identified in the time evolution graph sequence, risk(e) i ) represents event e i The risk value corresponding to the transaction behavior, where α and β are weighting balance coefficients.

[0131] Through a causal edge strength weighting mechanism, the system can not only identify causal relationships between transaction events, but also quantitatively express the strength of each causal edge based on historical frequency and risk impact. This weighting mechanism integrates the co-occurrence frequency of event pairs identified as causal edges in the historical graph sequence, as well as the risk value of the transaction behavior corresponding to the causal event. Through a two-factor weighted modeling approach, it effectively distinguishes between stable propagation paths and occasional interference paths, making the causal path graph more discriminative and interpretable. This mechanism provides a risk-logic-optimized edge weighting method for subsequent high-frequency causal chain screening and scoring graph construction, improving the structural expressiveness and decision rationality of causal reasoning in credit assessment scenarios.

[0132] S54. Calculate the cumulative weight values ​​of all causal edges in the causal path graph that are assigned causal strength weights, set a causal strength threshold, and extract causal edges whose cumulative weight values ​​are not lower than the causal strength threshold as high-frequency trading causal chains.

[0133] This invention extracts transaction event nodes that occur chronologically from a time-evolution graph sequence to construct a global time-series sequence of transaction events. Combining structural connectivity and behavioral attribute dependencies, it designs an event pair determination mechanism based on causal triggering conditions, constructs a causal path graph, and identifies causal chains with stable propagation characteristics. This enhances the system's ability to model potential risk propagation paths in complex transaction structures. Furthermore, while adding causal edges, a causal edge weighting mechanism is introduced. A causal edge weighting function is constructed by combining event co-occurrence frequency and risk level, achieving a quantitative expression of causal edge strength. This enables more accurate differentiation between key trigger points and edge paths in risk propagation. Finally, a threshold is set by the cumulative value of the causal edge weights in the causal path graph to filter high-frequency causal chains, significantly enhancing the ability to identify causal patterns in high-risk transactions. This provides clear, stable, and quantifiable causal evidence for scoring graph modeling and risk interpretation.

[0134] In this embodiment, S6 specifically includes:

[0135] S61. The final embedded representation of user nodes and transaction event nodes. The statistical feature vector c formed by the causal edge frequency and edge weight information in the associated high-frequency trading causal chain. i Perform joint modeling to generate joint feature representation z of graph nodes in the transaction scoring graph. i ;

[0136] S62, Based on joint feature representation z i Construct a transaction rating graph and calculate the value of each edge (v) in the transaction rating graph. i ,v j Risk association score r ij :

[0137]

[0138] Where, r ij ∈[0,1] represents the degree of risk association between graph node pairs, z i With z j Representing graph node v i With v j The joint feature representation, M is the trainable scoring weight matrix, w ij λ represents the causal edge weights between node pairs in the causal path graph, λ∈[0,1] is the causal weight adjustment coefficient, and σ(·) is the Sigmoid function.

[0139] Through a risk correlation calculation mechanism, the system comprehensively considers the joint feature representation of graph nodes and the edge weight information in the causal path graph when constructing a transaction scoring graph, establishing a quantitative model of risk propagation intensity. This mechanism achieves integrated modeling of structural information, semantic representation, and causal logic by jointly encoding the low-dimensional semantic representation of nodes and historical causal weights, and by introducing trainable scoring weight parameters and causal contribution adjustment factors. This avoids the bias problem of risk scoring relying solely on node representations or static relationships. This mechanism improves the consistency of edge connections in the scoring graph at both semantic and causal levels, effectively enhancing the ability to identify high-risk transaction behavior paths and providing stable support for risk visualization modeling and credit quantification assessment in vehicle transaction scenarios.

[0140] S63. Based on the risk correlation scores between user nodes and transaction event nodes in the rating graph, calculate the vehicle transaction risk coefficient and user credit score respectively, and output the risk level label of each transaction behavior and the credit score result of each user.

[0141] In this embodiment, S63 specifically includes:

[0142] S631. Based on the risk correlation score r between all user nodes and the target transaction event node in the transaction scoring graph. ij Calculate transaction event node v j Vehicle transaction risk coefficient Rj :

[0143]

[0144] in, Indicates the transaction event node v j There exists a set of all user nodes connected by the rating edge, r. ij Represents user node v i With transaction event node v j Risk association score, R j This represents the risk coefficient of a transaction event, with a value ranging from 0 to 1.

[0145] Through the vehicle transaction risk coefficient calculation mechanism, the system can quantify the overall risk level of a transaction based on the risk correlation scores between all user nodes and the target transaction event node in the transaction scoring graph. This mechanism constructs an event-oriented aggregated risk expression method by structurally aggregating the scoring contributions of all users connected to the target transaction node, overcoming the problem that traditional single-point scoring models cannot reflect risk consensus from a group perspective. Furthermore, the risk coefficient has a standardized interval expression form, facilitating comparative analysis and level classification across multiple transaction tasks, providing a clear and consistent quantitative basis for subsequent risk labeling and credit assessment reasoning.

[0146] S632. Based on the risk correlation score r between all transaction event nodes and target user nodes in the transaction scoring graph. ij Calculate user node v i User credit score C i :

[0147]

[0148] in, Indicates the relationship with user node v i There exists a set C of all transaction event nodes connected by a rating edge. i This represents the user's credit score, ranging from 0 to 1.

[0149] By employing a calculation mechanism for user credit scores, the system can inversely deduce a user's overall creditworthiness across multiple transaction scenarios based on the risk correlation scores between a user node and all its associated transaction event nodes in the transaction scoring graph. This mechanism uses the mean of all transaction risk scores and standardizes the numerical representation to ensure the comparability and interpretability of the scoring results on a unified scale. Through the integration of structured scoring with a broad range of transaction behaviors, this scoring mechanism effectively reflects the user's historical behavioral risk contribution, enhances the model's ability to assess the user's overall credit performance, and provides quantifiable support for credit rating, behavior tracking, and risk control linkage mechanisms.

[0150] S633, Based on the vehicle transaction risk coefficient R j And user credit score C i The risk level label is output for each transaction node, and the credit score is output for each user node. The risk level label is divided into three levels. When the risk coefficient R... j A value greater than 0.75 is marked as a high-risk level. j A risk level greater than 0.4 and less than or equal to 0.75 is designated as medium risk. j A value less than or equal to 0.4 is marked as a low-risk level.

[0151] This invention, in the process of modeling a transaction rating graph, jointly encodes the low-dimensional representations of user nodes and transaction event nodes with the frequency and risk information of causal edges in high-frequency transaction causal chains to generate joint feature representations of graph nodes, thereby improving the modeling accuracy of the coupling relationship between user behavior and transaction risk. Furthermore, by introducing a rating weight matrix and a causal edge weight fusion factor, a risk correlation function is constructed in the transaction rating graph, enabling the measurement of the risk propagation intensity between any pair of nodes in the graph and enhancing the semantic distinguishability of edge connections in the rating graph. Based on the rating result calculation, vehicle transaction risk coefficients are calculated for transaction event nodes, and user credit scores are calculated for user nodes. A multi-level risk level classification mechanism is designed, automatically labeling transaction levels according to risk value thresholds, and outputting structured credit evaluations and risk labels. This overcomes the technical bottleneck of traditional rating systems that cannot simultaneously reflect the risk source propagation path and the interpretability of rating level classifications, improving the comprehensive performance and operability of multi-objective decision-making evaluation in the vehicle transaction field.

[0152] Example:

[0153] To verify the feasibility and practical effectiveness of this invention, it was applied to the headquarters system of a large domestic used car trading platform in City B. This platform has an annual transaction volume exceeding 1.2 million vehicles and approximately 180,000 daily active users, encompassing multiple business modules including vehicle transactions, installment payments, insurance services, and dealer ratings. Due to the high amounts involved, the complexity of participating entities, and frequent cross-platform collaboration, the platform has long faced security risks such as insufficient user credit assessment, difficulty in identifying suspicious transaction behavior, and frequent vehicle transaction fraud. Especially with an imperfect user real-name authentication system, there is a large amount of abnormal behavior involving the manipulation of transaction ratings using multiple accounts, the repeated posting of false vehicle information, and the malicious creation of transaction chains. Traditional risk control systems struggle to trace these behavioral chains and the path to risk tracing.

[0154] The platform has decided to deploy the "Intelligent Early Warning and Credit Assessment Method for Vehicle Transaction Risk Based on Self-Evolving Heterogeneous Graph Neural Network and Credit Behavior Causal Inference Engine" proposed in this invention in its risk control architecture to improve the ability to model behavior from multi-source heterogeneous data and the accuracy of risk identification.

[0155] During deployment, the system first accessed nearly 12 months of historical data from the platform's transaction database, including vehicle transaction records, user behavior logs, third-party credit scores, and social interaction information from both inside and outside the platform. The system then performed unified labeling, time alignment, and feature normalization on this raw data to construct a raw transaction data set. Based on this set, a heterogeneous graph structure was built, containing user nodes, vehicle nodes, transaction event nodes, and social nodes. Furthermore, multiple types of edges were established based on relationships such as account similarity, social co-occurrence, and transaction behavior correlations, forming a heterogeneous graph relationship network.

[0156] To dynamically model the evolution of transaction behavior, the system slices the heterogeneous graph into weekly time windows, forming a weekly graph snapshot sequence. Based on the node addition rate and edge change rate in each graph snapshot, the system performs structural perturbation detection and adaptive graph structure reconstruction. Within the window where drastic structural changes are detected, the adjacency matrix and edge set are automatically reconstructed, forming a time-evolutionary graph sequence with dynamic structural evolution characteristics. Subsequently, based on the structurally updated graph snapshots, the system uses a graph neural network for node embedding propagation and employs a type-aware multi-layer aggregation mechanism to generate low-dimensional semantic representations of user and transaction event nodes.

[0157] The system further combines the transaction time sequence with behavioral causal dependencies to identify causal trigger paths between transaction events from the evolution graph and construct a causal path graph. For high-frequency stable path chains, the system assigns causal weights based on a joint calculation of frequency and risk, thereby selecting causal path subgraphs with high risk control value.

[0158] Finally, the system jointly models the semantic representations of user nodes and transaction nodes with causal subgraphs to generate a transaction rating graph. It then calculates the risk correlation score between each pair of nodes using a rating function, further calculates the risk coefficient of each transaction behavior and the credit score of each user, and automatically outputs the corresponding risk level label and user rating results.

[0159] Following the implementation of this system, the platform underwent a three-month practical evaluation, primarily testing three dimensions: risk identification capability, user credit score stability, and transaction fraud rate. The following is a comparison of the platform's key risk control indicators before and after deployment, detailed in Table 1:

[0160] Table 1: Comparison of Risk Control Performance Indicators of Used Car Platforms in City B

[0161]

[0162]

[0163] As can be seen from the table above, the deployment of this invention significantly improves the platform's ability to identify abnormal transactions. The system can identify an average of 129 high-risk transactions per month, approximately 2.7 times the number before deployment. Furthermore, because the causal path explanation mechanism supports more detailed risk causal analysis, the system issued automatic warnings for 312 abnormal transactions within three months, assisting the risk control team in early intervention. Ultimately, the coverage rate of confirmed and intercepted fraudulent activities increased from 41.2% to 87.6%.

[0164] Furthermore, the stability of the user credit scoring mechanism has been significantly improved, with the standard deviation decreasing from 0.61 to 0.23. This indicates that the scoring system is more consistent and stable, and is no longer easily affected by drastic fluctuations caused by transaction frequency or short-term behavior. Regarding abnormal account identification, the system utilizes multi-source information graph structures for joint modeling to identify operational similarities and behavioral co-occurrence relationships among multiple accounts. This improves the accuracy of account clustering identification from 65.7% to 93.2%, effectively combating malicious behavior on the platform where some accounts manipulate transaction scores using aliases.

[0165] After deployment, the system demonstrated excellent overall response efficiency, averaging 4.6 seconds in the complete process of comprehensive graph calculation, embedded propagation, and risk inference, meeting the platform's requirements for real-time response. Furthermore, after high-risk transactions were automatically flagged by the system, the manual confirmation response time was reduced from the original 42 minutes to 6 minutes, significantly alleviating the workload of manual risk control on the platform.

[0166] In a typical case, the system detected a user posting three transaction records for the same vehicle using different accounts within two weeks. Although the transaction amounts varied slightly, the timing, location, and social media account correlations were highly consistent. Through causal path graph comparison, the system determined this to be typical abnormal transaction behavior and issued a medium-risk warning in advance. After verification by the platform's risk control team, it was confirmed that the vehicle was illegally mortgaged, successfully preventing a loan transaction of 50,000 yuan.

[0167] In summary, the vehicle transaction risk assessment method proposed in this invention, based on a self-evolving heterogeneous graph neural network and a causal inference engine, can automatically identify structural changes from multidimensional data, uncover hidden causal chains, and provide interpretable risk warning results. In actual business, it significantly improves the platform's risk control capabilities, user identification accuracy, and response efficiency, verifying the effectiveness and practical value of this invention.

[0168] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent early warning and credit assessment of vehicle transaction risks based on big data analysis, characterized in that, Includes the following steps: S1. Collect vehicle transaction record data, user behavior trajectory data, third-party credit data and social interaction data, and perform identification normalization and time sequence alignment to generate the original transaction data set; S2. Construct a heterogeneous graph structure based on the original transaction data set; S3. Divide the heterogeneous graph structure into a sequence of graph snapshots according to a fixed time window, and perform adaptive structural reconstruction on the graph snapshots in each sequence based on the node addition rate and edge change rate to generate a time evolution graph sequence with time evolution characteristics. S3 specifically includes: S31. Divide the heterogeneous graph structure into time segments according to fixed time windows to obtain a sequence of graph snapshots. Among them, G t This represents the snapshot of the graph corresponding to the t-th time window, where n represents the total number of time windows; S32, For each graph snapshot G t Calculate the node addition rate α t The rate of change of the edge β t : Among them, V t and E t Representing the snapshot G respectively t The set of nodes and the set of edges, V t-1 With E t-1 These represent the set of nodes and the set of edges of the previous time window snapshot, respectively. The symbol \ indicates the set difference operation, and the symbol △ indicates the symmetric difference of sets. S33. When the following adaptive reconstruction condition is met: a t >θ v ∪β t >θ e ; Then for the snapshot G t Perform a structural update and update the graph snapshot G. t The adjacency matrix and node attribute representation, where θ v With θ e These are the thresholds for node addition rate and edge change rate, respectively. S34. Arrange all the graph snapshots that have undergone structural updates in chronological order to form a time evolution graph sequence. Among them, G' t Let m represent the graph snapshot after adaptive structural reconstruction, and m represent the number of updated graph snapshots, satisfying m≤n; S4. For each snapshot in the time evolution graph sequence, a graph neural network is used to perform node embedding calculations to generate low-dimensional feature representations of user nodes and transaction event nodes in each time window. S4 specifically includes: S41. For each of the time evolution graph snapshots G′ after adaptive structural reconstruction in the time evolution graph sequence. t User nodes and transaction event nodes are embedded and initialized separately, and initial representation vectors are set according to node types. S42. Before each propagation layer l begins, based on the distribution structure of heterogeneous node types in the graph snapshot, execute the type constraint sampling strategy for neighbor nodes to construct the type-aware neighbor set for each node in the propagation layer. S43. For each node v i Based on type-aware neighbor sets Construct the propagation input and collect the representations of neighboring nodes in the previous layer. And retain node type information; S44. Update node representations using a node type-aware aggregation mechanism: in, Represents node v in the l-th layer i The representation vector, Represents node v j The representation vector in the previous propagation layer l-1, Represents node v i In the set of neighboring nodes in the current graph snapshot, σ(·) represents a non-linear activation function, and MEAN(·) represents an aggregation function that performs an element-wise averaging operation on all input vectors in the set. This indicates that the node v i The trainable weight matrix corresponding to the node type φ(i) at the l-th layer, where φ(i) is the node type mapping function; S45, in the snapshot G′ t In each propagation layer l that performs graph neural network embedding computation, the node representation dimension and activation function type of the propagation layer are set, and the corresponding type propagation weight matrix in the propagation layer is configured for different types of graph nodes to control the representation update method of each type of graph node in the propagation layer. S46. After all propagation layers have been executed, extract the final representations of the user node and the transaction event node respectively. As a low-dimensional semantic representation of user nodes and transaction event nodes; S5. Based on the original transaction data set and time evolution sequence, construct a causal path graph according to the time sequence and behavioral dependencies of vehicle transaction record data, and extract high-frequency transaction causal chains from it; S6. Jointly model the low-dimensional feature representations of user nodes and transaction event nodes with the high-frequency transaction causal chain to generate a transaction scoring graph, and define a risk correlation function to calculate the vehicle transaction risk coefficient and user credit score. S5 specifically includes: S51, From the time evolution sequence Extract the transaction event nodes that occur within each time window, and combine them with the timestamps from the original transaction dataset to construct a global event sequence {e1, e2, ..., e...} for all transaction events in chronological order. n }, where e i This represents the i-th transaction event node, and n represents the total number of events; S52, in each graph snapshot G' after adaptive structural reconstruction t In the graph structure, based on the structural connections and behavioral attribute dependencies between transaction event nodes, it is determined whether event pairs in the global event sequence satisfy the causal triggering condition. If event e i Prior to e in time j And in G' t If structured behavioral interactions exist, then causal edges (e) are added to the causal path graph. i →e j ); S53, Adding causal edges (e i →e j At the same time, causal strength weights w are further assigned to the causal edges. ij w ij ∈[0,1]: w ij =α·freq(e i ,e j )+β·risk(e i ); Among them, freq(e i ,e j ) represents an event pair (e i ,e j The historical co-occurrence frequency of causal edges identified in the time evolution graph sequence, risk(e) i ) represents event e i The risk value corresponding to the transaction behavior, where α and β are weighting balance coefficients; S54. Calculate the cumulative weight values ​​of all causal edges in the causal path graph that are assigned causal strength weights, set a causal strength threshold, and extract causal edges whose cumulative weight values ​​are not lower than the causal strength threshold as high-frequency trading causal chains. S6 specifically includes: S61. The final embedded representation of user nodes and transaction event nodes. The statistical feature vector c formed by the causal edge frequency and edge weight information in the associated high-frequency trading causal chain. i Perform joint modeling to generate joint feature representation z of graph nodes in the transaction scoring graph. i ; S62, Based on joint feature representation z i Construct a transaction rating graph and calculate the value of each edge (v) in the transaction rating graph. i ,v j Risk association score r ij : Where, r ij ∈[0,1] represents the degree of risk association between graph node pairs, z i With z j Representing graph node v i With v j The joint feature representation, M is the trainable scoring weight matrix, w ij λ represents the causal edge weights between node pairs in the causal path graph, λ∈[0,1] is the causal weight adjustment coefficient, and σ(·) is the Sigmoid function. S63. Based on the risk correlation score between user nodes and transaction event nodes in the rating graph, calculate the vehicle transaction risk coefficient and user credit score respectively, and output the risk level label of each transaction behavior and the credit score result of each user. S7. Based on the transaction rating chart, output the risk level label of the target transaction and the corresponding user's credit score result.

2. The intelligent early warning and credit assessment method for vehicle transaction risks based on big data analysis according to claim 1, characterized in that, The original transaction data set includes vehicle transaction record data, user behavior trajectory data, third-party credit data, and social interaction data, which are then normalized and time-series aligned. The vehicle transaction record data is used to identify the transaction behavior relationship between users and vehicles, the user behavior trajectory data is used to identify the similarity relationship of account operations between different user accounts, and the social interaction data is used to mine the co-occurrence relationship of behaviors among users.

3. The intelligent early warning and credit assessment method for vehicle transaction risks based on big data analysis according to claim 1, characterized in that, The construction of the heterogeneous graph structure includes: constructing user nodes, vehicle nodes, transaction event nodes, and social nodes respectively using user entities, vehicle entities, transaction event entities, and social entities extracted from the original transaction data set; generating transaction behavior relationship edges between user nodes and vehicle nodes based on the mapping relationship between user identifiers and vehicle identifiers contained in the vehicle transaction record data; constructing account operation similarity relationship edges between different user accounts with similar account operation characteristics based on user behavior trajectory data; and generating social co-occurrence relationship edges between user pairs who jointly participate in transaction behavior based on message interaction behavior and such behavior in social interaction data, thus forming a heterogeneous graph structure containing multiple types of nodes and multiple types of relationship edges.

4. The intelligent early warning and credit assessment method for vehicle transaction risks based on big data analysis according to claim 1, characterized in that, Specifically, S33 includes: S331. For a graph snapshot G that satisfies the adaptive reconstruction condition... t All nodes v i Perform structural disturbance detection and compare the current neighbor set. Snapshot G from the previous image t-1 Neighbor set Constructing a set of perturbed neighbors S332, For snapshot G t Each node v i Based on the current node representation With the set of disturbed neighbor nodes The structure-aware weighted aggregation result is used to calculate the node representation after the structure update. Its expression is: Where γ∈[0,1] is the structural perturbation fusion factor. Indicates the disturbance of neighbor node v j At time t, the node is represented as follows: For node v j For node v i The perturbation structure affects the weights, satisfying: S333. Based on the node representation update result, reconstruct the graph snapshot G. t The adjacency matrix and edge set are used to retain the edge perturbation threshold θ whose change intensity exceeds the edge perturbation threshold. s Given the edge structure, generate a snapshot of the updated graph G′. t .

5. The intelligent early warning and credit assessment method for vehicle transaction risks based on big data analysis according to claim 1, characterized in that, S42 specifically includes: S421. In each graph snapshot G′ after adaptive structural reconstruction t In the middle, for each target graph node v i Obtain the type φ(i) of the node and extract the set of neighboring nodes that have connecting edges with the node in the graph. As a set of candidate neighbors; S422. Construct a type constraint matrix between node types. Where K is the total number of node types in the graph, Θ ab ∈[0,1] indicates that graph nodes of type a are allowed to sample graph nodes of type b; S423, Based on the target graph node v i The type φ(i) = a, for the set of candidate neighbor nodes Graph node v of type b j Assign sampling probability p ij ; S424, Based on the sampling probability p ij For candidate neighbor set Perform a sampling operation to obtain the target graph node v i Type-aware neighbor set in the current propagation layer l The number of sampling nodes is controlled by the hyperparameter d; S425. To improve the stability of the sampling process, multiple rounds of resampling are performed and the sampling frequency of candidate nodes is counted. If the frequency deviation exceeds a set threshold ∈, the sampling probability distribution p of the candidate neighbors is renormalized. ij .

6. The intelligent early warning and credit assessment method for vehicle transaction risks based on big data analysis according to claim 1, characterized in that, Specifically, S63 includes: S631. Based on the risk correlation score r between all user nodes and the target transaction event node in the transaction scoring graph. ij Calculate transaction event node v j Vehicle transaction risk coefficient R j : in, Indicates the transaction event node v j There exists a set of all user nodes connected by the rating edge, r. ij Represents user node v i With transaction event node v j Risk association score, R j This represents the risk coefficient of a transaction event, with a value ranging from 0 to 1. S632. Based on the risk correlation score r between all transaction event nodes and target user nodes in the transaction scoring graph. ij Calculate user node v i User credit score C i : in, Indicates the relationship with user node v i There exists a set C of all transaction event nodes connected by a rating edge. i This represents the user's credit score, ranging from 0 to 1. S633, Based on the vehicle transaction risk coefficient R j And user credit score C i The risk level label is output for each transaction node, and the credit score is output for each user node. The risk level label is divided into three levels. When the risk coefficient R... j A value greater than 0.75 is marked as a high-risk level. j A risk level greater than 0.4 and less than or equal to 0.75 is designated as medium risk. j A value less than or equal to 0.4 is marked as a low-risk level.

Images